Keyness

Keyness is a term used in corpus linguistics to describe how characteristic a word or lexical item is of a target corpus relative to a reference corpus. A word is said to have high keyness if it occurs more frequently in the target corpus than would be expected given the reference distribution; conversely, a word with low or negative keyness is underrepresented in the target corpus.

Calculation typically involves comparing observed word frequencies in the target and reference corpora with an expected distribution. The standard approach uses a 2x2 contingency table: word present or absent versus target or reference corpus. Statistical measures such as the log-likelihood ratio (G-test) or Pearson chi-square quantify the deviation from expectation. Other variants include standardized log odds of frequency, log odds ratio, and permutation-based methods. The result is a keyness score, often used to rank words to form a keyword list for the target corpus.

Keyness analysis is widely used for exploratory discourse analysis, genre description, authorial studies, historical linguistics, and media analysis. It helps identify terms that signal topics, styles, or sociolinguistic features distinctive of the target corpus. Positive keyness highlights items that define the material, while negative keyness highlights items relatively suppressed.

Limitations and cautions include dependence on the quality and comparability of the corpora, sample size effects, and genre or topic differences that may drive results rather than intrinsic characteristics. Statistical significance does not in itself determine interpretive importance; qualitative analysis is usually required to explain why a term is keyworded in a given context.